GTX 690

[awsjolie]

I've just get a geforce GTX 690 and tried it with 2.58e Kepler build but its really slow for my 200k poly or over scenes...

Anybody know that when will Octane support totally for kepler card???? It's been a long time since the first kepler appeared on the market

[gueoct]

Come on, it´s been 3 month...
Since then, this topic has been chewed through, and otoy team was
totally honest about their progress.

[Stefan42]

I am also using a GTX 690 card as you, bought it 1 week ago just to be able to use iRay, Arion and Octane with more CUDA cores so I can't wait for Octane to have this feature as well. Hope/think they are working with it, but I guess it takes time, and who knows, by the time they are done, a new GTX 7xx card is out that is even better than GTX 690

[acc24ex]

so iRay and Arion are performing better with kepler?
- maybe there's some programming conspiracy going on there
- could be a that it will perform better than anything on the next generation of gpus, I know as soon as 6xx came out, there was talk that nvidia was holding back the perfomance
- they sort of did it with gtx480 - nvidia released it until 580 was ready, - 15% slower and 50% hotter

[abstrax]

As I have written before, we plan to release a version this week that has similar Fermi performance as before and better Kepler performance than previous Kepler builds. Unfortunately the Keplers are still 10-20% slower than comparable Fermi cards. The main reason for that is that Keplers have 192 cores per multi-processor and only 8 multi-processors, while Fermis have 32 cores per multi-processor and 16 multi-processors. Having more multi-processors is a good thing, having many cores per multi-processor is a bad thing, since it requires your code to run more coherent, which is hard to achieve.

To give you a rough idea: To fully utilize the cores of a Kepler multi-processor, you need to be able to run 192x the same instruction on different data and should not branch. Now, there are some mechanisms on CUDA GPUs that help reducing the problem, but it's problematic nonetheless. And that's the reason why you don't achieve a speed gain with Keplers. Pure ray-tracing (which runs more coherent than say shading) actually runs faster on Keplers than on Fermis, which you can see when you run the deep channel kernel, but as soon as you add the full render pipeline, Kepler don't shine so much anymore.

That's the current situation. We still plan to do some performance work in the future, but there are also gazillions of feature requests and other stuff to do, so in the end it's all about priorities...

Cheers,
Marcus

[awsjolie]

Thank Marcus, so ill keep waiting

I did really enjoy working with Arion 1.6, it was my best choice in past although there're bugs around and hope that one day theyll release a better version . But now their new maxlive 2.0 require more complex operations with unfriendly interfaces and very very bad performance with kepler cards, which i cant get use to...

From now on i consider that Octane could be the best in the near future, so now its my choice.

Thank you for supporting. In my opinion you're leading

Cheers!

[gabrielefx]

@Marcus

I don't understand if Maximus on Quadro + Teslas will perform better than using multiple GTXs

I would like to know if Otoy will consider to test and tune the new Octane releases (cuda 5 based) with the new professional Nvidia boards.

I think that Otoy should stipulate agreements with Nvidia.

I see videos of Adobe, Autodesk and Dassault and Chaos Group too... why not Otoy?

Octane is the faster gpu rendering engine on the planet and produces photorealistic results better than whatever render engine we use.

Nvidia should be interested in your research and development. Nvidia should help you..

Thanks to Otoy Nvidia could sell more Teslas and make more money.

[abstrax]

@Marcus

I don't understand if Maximus on Quadro + Teslas will perform better than using multiple GTXs

I would like to know if Otoy will consider to test and tune the new Octane releases (cuda 5 based) with the new professional Nvidia boards.

I think that Otoy should stipulate agreements with Nvidia.

I see videos of Adobe, Autodesk and Dassault and Chaos Group too... why not Otoy?

Octane is the faster gpu rendering engine on the planet and produces photorealistic results better than whatever render engine we use.

Nvidia should be interested in your research and development. Nvidia should help you..

Thanks to Otoy Nvidia could sell more Teslas and make more money.

Well, I guess Octane is some kind of competition to NVIDIA's iRay

Marcus

[gueoct]

actually, it´s no competition....

[mbetke]

But even the Nvida iray team guys say there will not be same kepler performance like on fermi.
And the new Kepler pro-boards are for double-precision only.

the bad thing is that it seems you guys have to re-code half your renderer if nvidia decides to change something (to their advantage).